Radar ·
Survey maps self-evolving coding agents
A survey analyzes 108 systems that learn from software development interactions and use those experiences to change memory, workflows, tools, or model parameters.
The survey examines 108 coding-agent systems that learn from software development interactions. It focuses on how prior experiences can guide changes to the agents themselves.
The survey maps adaptations to memory, workflows, tools, and model parameters. For teams studying or applying these ideas with AI, protecting organizational data still matters: avoid sending confidential information and use appropriate access controls.